Abstract
This work explores the dimension reduction problem for Bayesian nonparametric regression and density estimation. More precisely, we are interested in estimating a functional parameter f over the unit ball in Rd, which depends only on a d∗-dimensional subspace of Rd, with d∗<d. It is well-known that rescaled Gaussian process priors over a given function space achieve smoothness adaptation and posterior contraction with near minimax-optimal rates. Furthermore, hierarchical extensions of this approach, equipped with subspace projection, can also adapt to the intrinsic dimension d∗ (Tokdar (2011)). When the ambient dimension d does not vary with n, the minimax rate remains of the order n−β∕(2β+d∗), where β denotes the smoothness of f. However, this is up to multiplicative constants that can become prohibitively large as d increases. The dependence between the contraction rate and the ambient dimension has not been fully explored yet and this work provides a first insight: we let the dimensions d∗ and d grow with n, and by combining the arguments of Tokdar (2011) and Castillo and Randrianarisoa (2024), we derive growth rates for them that still lead to posterior consistency with minimax rate. We also discuss the optimality of the growth rate for d. Additionally, we provide a set of assumptions under which a consistent estimation of f leads to correct estimation of the subspace projection, assuming that d∗ is known.

This publication has 32 references indexed in Scilit: